What an AI can and cannot see
Most of what is on an issue is readable by an assistant. The gaps are specific, and each one is the kind of thing you discover after twenty minutes of wondering why the assistant is being obtuse.
Images: mostly yes, with two gaps
An assistant reading an issue's comments now gets the pictures as well as the text - both the screenshots attached to comments and the ones that arrived as viewpoints from another BIM tool. So "look at this issue and tell me what the problem is" works even when the problem is in a photograph.
Two things it still will not see:
Files attached to the issue itself, rather than to one of its comments, are not returned. In practice most attachments end up on a comment anyway - see Description, files and images - but if something important is on the issue, say so in words too.
Long threads get truncated. There is a cap on how many images come back in one response, and very large ones are skipped. The assistant is told which ones it did not get, so it can say so rather than guess.
The order of cards on the board: now visible
This used to be invisible. It is not any more.
A card's place in its column is stored on the issue, and an assistant can both read it and change it. It can answer "what is at the top of the list?" and it can reorder for you.
One thing to know: reordering is not recorded anywhere. It does not appear in the issue's history or in the activity feed, and it sends no email. So if a card has moved and you want to know who moved it, nobody can tell you - assistant or not. That is true whether a person or an assistant did it.
It only accepts the values your project allows
Ask for a status your project does not have and the assistant will be refused, told the real list, and told to ask you rather than substitute something.
That is working as intended. An assistant that quietly picked the nearest-looking value would fill the tracker with vocabulary nobody chose.
What to do: either use one of the listed values, or add the one you want under Project options. The fix is in the project settings, not in the assistant.
One convenience worth knowing: an issue an assistant creates without a status gets On hold, rather than being left blank. It is the least presumptuous of the four, and it means new issues do not fall out of the board.
Also worth knowing
It cannot delete anything - see Connecting an AI assistant.
It cannot add people to a project. Adding somebody is a decision a person makes in the app. An assistant can assign an issue to anybody already on the project - including placeholder people - but if the person you name is not there, it will say so and leave the issue unassigned rather than guessing.
There is a limit on how much it can ask for. A connected assistant gets a budget of requests per hour. Ordinary use is nowhere near it; a runaway loop hits it and is told to wait, which is the point.
It may be signed in as a different account than you expect, if the connector was set up with another login. When in doubt, ask it who it is signed in as.
It sees the standard's words, not the app's. It will talk about topics and stages where you say issues and milestones. See Issue, task or topic?, or just ask it to use your words.
What it is genuinely good at
Reading across many issues and telling you what is there: what is assigned to you and overdue, which issues have gone quiet, what changed on something this week, summarising a long comment thread - and now, describing what is in the screenshots attached to it.